Symbiosis as a Metaphor (2021)
About
Program notes
Symbiosis as a Metaphor is inspired by the concept of the “black box of AI”. The concept refers to the part of the learning process of a neural network that is not fully transparent to the human.
The piece utilises a NN that is trained on given sound material and tries to generate similar material. The learning process of AI is metaphorised as a symbiotic struggle between the human expectation, represented by the original sound, and the AI’s attempt to interpret and
mimic it, represented by the generated sound. This dipole is reflected through two audiovisual objects.
From a diegetic perspective, these two objects operate like organisms, developing behavioural synchretic relations. On one hand, the organic, fluid and complex organism, representing the idealised outcome, the original sound. On the other hand, the reduced and skeletonised version of it, representing the AI’s mimicry. The conflict and co-existence between these two sets the stage for the narrative of the piece.
Technical description
The NN used for this piece is PRiSM-SampleRNN [55], a variation of the original SampleRNN model, designed by Christopher Melen for the RNCM PRiSM (Royal Northern College of Music Centre for Practice and Research in Science and Music). This algorithm is trained on time-series data and generates predictions based on its training. For this work, the network was trained on a 90-minute soundset created through a generative patch on a modular synthesiser.
If the training is "successful", PRiSM-SampleRNN is expected to generate audio that sounds similar, to some degree, to the input sound. During training, PRiSM-SampleRNN generated sound samples demonstrating its progress. These samples were used as structural pillars in the course of the piece, designating the evolution of the learning process of the NN. Together with the generated material, the piece includes the original soundset that was used as input data. The two soundworlds, the original and the generated one, are intertwined and juxtaposed throughout the entire work.
I would like to thank Christopher Melen for his contribution to the training of PRiSM-SampleRNN as well as the insightful discussions.
Format: Fixed audiovisual media
Duration: 10:46
Performances/Exhibitions
16/06/2021 - Future Music #3 - UnSupervised, broadcasted online
27/11/2021 - Manchester Theater in Sound Festival, at Cosmo Rodewald Hall, Manchester, UK
14/05/2023 - Athens Digital Arts Festival “Welcome to the Uncanny”, at Astor Cinema, Athens, Greece
24/09/2024 - Musicacoustica-Hangzhou 2024, Hangzhou, China